Explore the tools I use to turn research questions into experiments, analyses, and interactive systems.
Benchmarking Large Language Models and Computer Vision for Automated Construction Waste Recognition & The Impacts of Inputting Data on Visual-Language Models in Automated Construction Waste Recognition (HKU)
Primarily utilized API techniques, prompt engineering, and few-shot techniques to evaluate the potential of LLM/LVM in image recognition tasks.
Developing a LLM for Social Science Research (Oxford)
Primarily employed fine-tuning and RAG on social science databases (WVS and ESS mapping), comparing the trained model against its default aligned equivalent.
Official Media as Emotional Valves: How Official Media Guides Nationalism on Chinese Social Media
Contributed to research by utilizing supervised text classification and topic models, along with BERT-based sentiment analysis, evaluating the government’s responsiveness to citizen complaints during the COVID-19 pandemic.
Topics and Sentiment Analysis of International Communication Related to Chinese Handicrafts on TikTok
As an early adopter of data collection via the TikTok API and web scraping techniques, I conducted research in 2022 when academic studies on the TikTok platform were still scarce. I employed the VADER and LDA models to investigate sentiment polarization and dynamic topic distribution.
Benchmarking Large Language Models and Computer Vision for Automated Construction Waste Recognition
Benchmarked computer vision models trained with CNN, Attention, and Transformer on 500 labeled waste disposal images, achieving model accuracy above 90.40%.
China Artificial Intelligence and Future Media Innovation and Entrepreneurship Competition
Created a smart health monitoring system with wearable health units using foot pressure and heart rate analysis models based on MobileNet and CNN deep learning models, secured a grant of $1,400.
Hong Kong Housing Authority Android App: e-InStar
Independently developed an app using the Flutter framework, deployed for practical use by over 10,000 users in Hong Kong’s engineering industry. The app implements NFC communication, QR code scanning for uploading on-site report images and engineering reports, and real-time collaboration features for multiple user groups to review project progress based on uploaded images.
Mitigating or Exacerbating Polarization? An Intervention Study on Alleviating Online Misogyny Through Generative AI Dialogues
Completed a Master’s dissertation developing a conversational AI model and integrating an Internalized Misogyny Scale to measure participants’ misogyny levels through an online controlled experiment. Created an AI chat assistant using the GPT-4 API within a Flutter app, hosted on Tencent Cloud with Firebase for data storage, aimed at mitigating misogyny in conversations.
National First Prize(1st/800 teams)2021 China Data Journalism Competition:They Tried to Melt an Iceberg
This project is a self-developed and coded(with HTML/CSS/JavaScript/D3.js) data journalism piece that combines
dynamic visualization with in-depth reporting to bring to light the situation of rare diseases, much like an iceberg.
The output of this project can be viewed at RareDisease .